
founder @usenorilu ship self-healing agents french prononciation @spokira previously sandboxes, webrtc, real-time transcription and inference pipelines
Paris, France
Joined June 2018
- Tweets3.9K
- Following1.4K
- Followers3.4K
- Likes27.9K
Pinned Tweet
Hey @FlutterDev! Just published my first article on “How to call a Rust function from Dart using FFI” medium.com/@Sach97/how-to-ca…
Sacha Arbonel retweeted
I know I'm late to the party, but I was busy shipping and vacationing.
I am a proud member of the @HyperFrames_ hype club, and I wanted to share the behind-the-scenes of one of my latest videos that I edited with HyperFrames and @claudeai
I've been agentically editing and visualising videos for a very, very long time now. Way before the hype picked up. FFmpeg is like the bane of my existence. I also built multiple skills to edit videos.
But I think Hyperframes is what Claude Code is to Opus, Codex is to Sol, livekit / pipecat is to Gradium.
Shared more about how I used it in the video.
PS: I still like to keep my videos super raw, mostly. Too much dopamine ain't good for ya'll
80K+ downloads and counting for Phonon.
Here are @neilzegh and @lmazare answering some questions from the community.
gradium.link/phonon
Always feel good when the CEO of a company making 80 millions ARR takes the time to answer your LinkedIn message
Even if it's his bot
ICP seems good now I need to do more volume across cold emailing and LinkedIn
Cold emailing is sorted my inboxes are warming up
Good advices for LinkedIn anyone ?
4
209
At this point I don't know how boardy makes money and I'm too afraid to ask
But good recos so far
1
7
393
That ahah moment when you see an indirect competitor (with funding) with a copy and design so clean that you want to take inspiration from it
And you realise I need to step up my design and copy
Like it just clicks and you realise someone stepped into what you are building and validating your market
Frontier models being bad at abstractions is beyond me
They need so much hand holding it's tiring
It usually take 2 or 3 iterations to get to something usable
That is slop. Frontend slop is acceptable imho easily fixable
Abstraction slop compounds (breaking changes, conversion, etc)
2
5
362
I think at this point 50 pct of my ram is dedicated to computer use / browser use for my agents to verify their work
1
1
272
Astra terra loves to make green frontends
Also my neobank branding just switched to green out of the blue
should I be worried
189
Just started using Granola for a discovery call.
Not a great start. Not giving access to the audio is so stupid
1
3
321
I don't get people that says cold emailing does not work
I got 1 answer out of 5 emails sent
Much better than linkedin outreach so far
1
3
349
Made a vectorization plugin to test the waters of Openai plugin submission process
Works pretty well
comment "mallace" and I'll send you a link to the plugin
3
207
Can someone at OpenAI use that to teach Astra opus level of taste
Thanks. I just want one subscription
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
1
185
The phone call scammers using voice ai are getting pretty good
now they hesitate and have subtle regional accents and what not
There are still dead giveaways but scary stuff
1
175








